Persistent memory for AI agents.
AI Agent Tooling
Mindroot is a local-first persistent memory system for AI agents, combining embedding-searched memories with human-editable Markdown notes through HTTP, MCP, CLI, and dashboard interfaces.
AI agents lose valuable project knowledge across transient sessions, forcing teams to repeatedly rediscover architecture, conventions, commands, and recurring implementation gotchas.
The memory layer needed semantic and keyword retrieval without turning durable knowledge into an opaque database, while keeping every result strictly scoped to the correct project.
We built Mindroot around a two-layer model: concise, retrieval-optimized memories surface the right context quickly, while human-editable Markdown notes remain the durable source of truth.
A local Node.js service combines SQLite FTS5, ONNX embeddings, project-scoped MCP tools, an HTTP API, CLI commands, and a visual dashboard so agents and people can maintain the same transparent knowledge base.